Semiconductor metrology system and method

An optical critical dimension method comprising:

• performing optical measurement of an associated wafer by an associated metrology tool, the measurement including a plurality of measured optical spectra;
• using a machine learning (ML) model to determine a critical dimension inference from the plurality of measured optical spectra, the ML learning model trained using a training set including a plurality of training optical spectra;
• generating a plurality of theoretical optical spectra for the associated wafer by a machine learning predictor component in accordance with the training set;
• comparing the measured optical spectra to the theoretical spectra;
• evaluating a fitting of the measured optical spectra to the theoretical optical spectra;
• and
• generating an acceptance indicator for the calculated critical dimension inference for the associated wafer responsive to an output of the evaluation,
• wherein the ML learning model is trained by a training method comprising:
• retrieving, from at least one associated database, the training set including the plurality of training optical spectra acquired from a plurality of reference wafers and the training set including a corresponding plurality of references indicative of critical dimension parameters for the plurality of reference wafers;
• extracting, from the training set, a plurality of features;
• performing regression analysis on the plurality of extracted features;
• generating predictor variables for a machine learning algorithm associated with the ML learning model in accordance with an output of the regression analysis on the plurality of extracted features;
• generating a machine learning optical critical dimension explainer based on a mask and the plurality of extracted features from the training set, and combining the mask with generated inverse feature functions based on the plurality of extracted features;
• and
• outputting, by the machine learning optical critical dimension explainer, a visual representation of optical critical dimension measurements on an associated display,
• wherein at least one of the retrieving, extracting, performing, calculating, generating, evaluating, and outputting is performed by a processor in communication with memory.

An optical critical dimension method comprising:

• performing optical measurement of an associated wafer by an associated metrology tool, the measurement including a plurality of measured spectra;
• using a machine learning (ML) model to determine a critical dimension inference from the plurality of measured spectra, the ML learning model trained using a training set including a plurality of training spectra;
• comparing the measured spectra to reference spectra data generated from the training set;
• generating a machine-learning optical critical dimension explainer based on a mask and one or more inverse feature functions extracted from the training set, applying the generated inverse feature functions to generated predictor variables based on performing a regression analysis on the extracted plurality of inverse feature functions subsequent to application of the mask to the generated predictor variable, and combining the applied inverse feature functions with the mask and outputting, by the machine-learning optical critical dimension explainer, a visual representation of optical critical dimension measurements on an associated display;
• and
• generating an acceptance indicator for the calculated critical dimension inference for the associated wafer responsive to an output of the evaluation,
• receiving a modification to the critical dimensionality of the measured spectra;
• simulating the modification in critical dimensionality of the measured spectra in response to an output of the evaluation by inputting into the ML-OCD explainer the measured spectra and the received change in critical dimensionality;
• and
• outputting a fitting quality of the simulated change in critical dimensionality on an associated display,
• wherein at least one of the performing, calculating, generating, evaluating, and generating is performed by a processor in communication with memory.